Paper Title
INSURANCE RENEWAL FORECAST

Abstract
Abstract -This study aims to correctly predict the renewal rates of insurance products, in particular vehicle insurance products, using Machine Learning (ML) methods and VDF Insurance Services’ customer data as a source. Customer identifying information such as age, current location, and vehicle information such as vehicle age and vehicle type will be used alongside with the customer past habitual data such as renewal time and renewal zone. After obtaining datasets, various ML methods were tried including Random Forest, Xgboost, Cat Boost, NGBoost and Light GBM to build the most accurate algorithm. In the study, individual and corporate customers were handled separately. The outcome of the algorithm is anumerical value ranging from 0 to 1, or the percentage of successful predictions. Results higher than 0.5 are treated as positive, meaning the customer will renew the insurance and the rest is treated as negative, meaning the customer will not renew. It has been found that among the ML methods Light GBM was the most accurate one with an F1 score of 0.83 for individual customers and 0.70 for corporate customers. It is evaluated by the company that the insurance renewal rates will increase by 15% if the prediction methods developed in this study are used before the renewal process.